[No Code] More models for braindecode

[robintibor]

One could have a look at the methods here, and test which ones improve EEG decoding, e.g., on High-Gamma Dataset/BCI Competition IV 2a/TUH Abnormal
@Div12345
https://pytorch.org/blog/how-to-train-state-of-the-art-models-using-torchvision-latest-primitives/

[bruAristimunha]

Is it better to make a list of possible architectures to implement in braindecode? @Div12345, do you have another suggestion?

[bruAristimunha]

Do you have some models in mind @robintibor and @agramfort? I am interested in this task, maybe could be a nice task to build the braindecode paper.

My list:

• LDMA
• All the models inside the BioT
• TFC
• MAEEG
• Dist-dcrnn
• EEGClip
• Bendr
• XSleepNet
• EEGGan
• EEGViT
• Embedding EEG

[robintibor]

Interesting to me could be:

• augmentations, especially channel dropout
• random erasing
• label smoothing
• mipup/cutmix
• EMA or SWA (Stochastic Weight Averaging #239)

[robintibor]

also an EEG version of TrivialAugment could be interesting. That would mean selecting a subset of augmentations, creating a discrete set of strengths per augmentation and selecting a random augmentation with a random strength for each example each time. I guess we would need to ensure that all the augmentations are likely to preserve the class label.

[bruAristimunha]

The neurips papers that I found interesting and maybe linked with braindecode.

https://neurips.cc/virtual/2023/poster/70774
https://neurips.cc/virtual/2023/poster/71605
https://neurips.cc/virtual/2023/poster/71107
https://neurips.cc/virtual/2023/poster/70217
https://neurips.cc/virtual/2023/poster/72228
https://neurips.cc/virtual/2023/poster/72383
https://neurips.cc/virtual/2023/poster/71117
https://neurips.cc/virtual/2023/poster/72408

[robintibor]

wow super interesting thanks @bruAristimunha

[bruAristimunha]

One more: https://github.com/duanyiqun/DeWave